[U-Boot] [PATCH] fs: fat: read: fix fat16 ls/read issue

[Simon Glass]

Hi Przemyslaw,

On 12 December 2014 at 08:30, Przemyslaw Marczak <p.marczak at samsung.com> wrote:
> Hello,
>
>
> On 12/12/2014 01:32 AM, Simon Glass wrote:
>>
>> Hi Przemyslaw,
>>
>> On 11 December 2014 at 05:01, Przemyslaw Marczak <p.marczak at samsung.com>
>> wrote:
>>>
>>>
>>> The present fat implementation ignores FAT16 long name
>>> directory entries which aren't placed in a single sector.
>>>
>>> This was becouse of the buffer was always filled by the
>>> two sectors, and the loop was made also for two sectors.
>>>
>>> If some file long name entries are stored in two sectors,
>>> the we have two cases:
>>>
>>> Case 1:
>>> Both of sectors are in the buffer - all required data
>>> for long file name is in the buffer.
>>> - Read OK!
>>>
>>> Case 2:
>>> The current directory entry is placed at the end of the
>>> second buffered sector. And the next entries are placed
>>> in a sector which is not buffered yet. Then two next
>>> sectors are buffered and the mentioned entry is ignored.
>>> - Read fail!
>>>
>>> This commit fixes this issue by:
>>> - read two sectors after loop on each single is done
>>> - keep the last used sector as a first in the buffer
>>>    before the read of two next
>>>
>>> The commit doesn't affects the fat32 imlementation,
>>> which works good as previous.
>>
>>
>>
>> This is very interesting\! Is this the same failure that I saw on this
>> thread?
>>
>>
>> http://u-boot.10912.n7.nabble.com/PATCH-U-Boot-ARM-rpi-b-detect-board-revision-td196720.html
>>
>> (search for fatload)
>>
>> "I tried this out. It worked OK for me except that it can't find the
>> device tree file bcm2835-rpi-b-rev2.dtb.
>>
>> Oddly I can fatload it from /bcm2835-rpi-b-rev2.dtb but when I try
>> from /syslinux/..//bcm2835-rpi-b-rev2.dtb it fails and cannot find the
>> file. Reducing the filename length to 8 chars works. I wonder what
>> year of my life FAT will stop plaguing me? "
>>
>>
>> Also can you write a test for this in test/fs/fs-test.sh?
>>
>> Regards,
>> Simon
>>
>> [snip]
>>
>
> Probably this is an another case which is caused by the sector buffer bug.
> Does this patch fixed your issue?
>
> I have some simple test for manual use with the ums tool.
> It just copy the test files to the tested fat16 partition mounted using the
> UMS, next it computes CRC32 of those files on the host and next using U-Boot
> fatload/crc32 commands - it tests the read feature. But it's not full
> automated - I didn't work on getting the log from U-Boot console.
>
> So I could check if the file checksums are proper and if all files were
> found on the partiion, by the U-Boot read command. It's not useful for the
> "test/fs/fs-test.sh" because this is not designed for the sandbox.
> My test writes some commands directly to U-Boot console, like this: echo
> "some cmd" > /dev/ttyS0.
>
> Unfortunately the sandbox config seems to be broken.
>
> The bug was not so obvious, any read/write on fat partition can change fat
> directory entries or add the new ones and then all data can be read right.
>
> I will send the scripts for such simple test.

I'm not sure if it fixes my problem but it seems likely. I will see if
I can make time to test it.

If you want to write input data to U-Boot sandbox we can do that
fairly easily. You can import cros_subprocess and use the function
there to generate output from your test and inspect the input from
U-Boot's command line. Let me know if you'd like an example.

Regards,
Simon